← Back to community benchmarks
gemma-4-31B-it-PARO
Performance
8k
tokens
110.3
PP tok/s
12.4
TG tok/s
74258
TTFT (ms)
20.0
Peak mem (GB)
Hardware
Chip
M1 Max (32c)
Memory
64 GB
GPU Cores
32
Software
oMLX
v0.3.9
macOS
macOS 26.5
Context
8,192
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 113.8 | 14.7 | 17.0 GB | view |
| 4k | 110.8 | 13.9 | 19.5 GB | view |
| 8k | 110.3 | 12.4 | 20.0 GB | current |
| 16k | 105.9 | 11.7 | 21.7 GB | view |
| 32k | 97.2 | 9.4 | 24.1 GB | view |
| 64k | 85.2 | 6.6 | 30.1 GB | view |
| 128k | 69.0 | 4.5 | 42.2 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 14.7 | 1.00× |
| 2× | 19.4 | 1.32× |
| 4× | 21.7 | 1.48× |